Maximizing Cloud Resource Utility: Region-Adaptive Optimization via Machine Learning-Informed Spot Price Predictions

EasyChair Preprint no. 11658

Abstract

This research paper presents a comprehensive study on the use of machine learn-ing models for price prediction of spot instances in various geographic regions in Amazon Web Services (AWS). The work focuses on forecasting prices across eleven unique locations using XGBoost and RandomForestregressors, with the goal of revealing significant insights into pricing dynamics and prediction accu-racy. The research explores how well these models anticipate prices, finds factors that influence price fluctuation, and assesses the practical consequences of these predictions for enterprises. The study employs a dataset containing pricing data from several places in a methodical manner. The study's findings reveal noteworthy trends and findings. The predicted performance of the models varies by region providing for region-specific insights into price forecast accuracy. To measure prediction perfor-mance, models are evaluated using Mean Squared Error (MSE) and Mean Abso-lute Error (MAE). Significantly accurate forecasts show that the models can successfully capture pricing changes. A comparison of the XGBoost and RandomForest models also offers light on their relative performance, which will benefit in algorithm selec-tion for future investigations.

Keyphrases: AWS, price prediction, Random Forest Regressor, Spot Instances, XGBoost

@Booklet{EasyChair:11658,
  author = {Kavita Srivastava and Manisha Agarwal},
  title = {Maximizing Cloud Resource Utility: Region-Adaptive Optimization via Machine Learning-Informed Spot Price Predictions},
  howpublished = {EasyChair Preprint no. 11658},

  year = {EasyChair, 2024}}